Neural networks for model predictive control

This paper is focused on developing a model predictive control (MPC) based on recurrent neural network (NN) models. Two regression NN models suitable for prediction purposes are proposed. In order to reduce their computational complexity and to improve their prediction ability, issues related with optimal NN structure (lag space selection, number of hidden nodes), pruning techniques and identification strategies are discussed. The NN-based MPC and the traditional PI (Proportional-Integral) control are tested in the presence of process disturbances on a crystallizer dynamic simulator.